Phase 2: Intent detection for session mode selection

[tessak22]

What

After 6-8 weeks of real usage data, add an intent detection layer so users can describe what they want to do in natural language and the app auto-selects the correct session mode and artifact type.

Scope

• Text input on dashboard: "What do you want to work on?"
• AI classifies intent into one of the 5 modes + artifact type (for create mode)
• Pre-selects mode and artifact type with a confirmation step ("I think you want a Create session for email_sequence — correct?")
• Explicit mode selection remains as permanent override
• Train on historical session data: what modes users actually chose for similar prompts

Reference

SPEC.md Section 5 — Skill Routing, Phase 2. Explicitly out of scope for v1, designed for after real usage validates the mode taxonomy.